Towards Autonomous Analysis in Finance: A Modular and Extensible Architecture for LLM-Based AI Agents

The complexity and dynamics of financial markets make it difficult for investors and researchers to make decisions. In this regard, this research proposes an AI-based Financial Data Analysis and Reporting Assistant that incorporates technical analysis, fundamental analysis, and news-based sentiment analysis for stock and exchange-traded fund (ETF) analysis. Built with real-time information from Google Gemini 2.0 Flash, it uses a three-node structure based on LangGraph (Chatbot Node, Tools Node, and Human Node) and Yahoo Finance API to create user-directed reports with a platform that supports conversation mode. Core components of this system include caching of data, technical analysis elements (RSI, MACD, Bollinger Bands), key financial ratios, news analysis with BeautifulSoup for news-associated information, and data representation with Plotly graphs. The performance of this system for AAPL and TUPRS.IS stocks show its ability to correctly operate in global and local financial markets with proper upholding of ethical principles by not giving investment recommendations. The findings highlight the system's ability to process multilingual data, mitigate hallucinations through context awareness, and provide holistic financial decision support. Limitations include dependence on third-party APIs and potential data delays. The work contributes to the field of financial AI by filling gaps in integrated and conversational systems and provides a modular framework for future predictive analytics and multi-asset portfolio development.